time series design

Definition
Time series design: A type of quasi-experimental research design that attempts to establish control via multiple observations of one group before and after treatment.

Notes
Sometimes a comparison group is simply not feasible for a variety of reasons such as cost, ethics, practicality, and so forth. In the absence of a comparison group, a design known as a time series quasi-experiment can be used that derives its control from observations over time instead of the comparison of one group against another (Suter 2012, pg. 23)

A time series design is shown below:
Pre Pre Pre Pre Pre T Post Post Post Post Post

In this design, the object is to link a break in the trend revealed over time to the introduction of the treatment. The break in the trend should occur at the same time as or shortly after the treatment introduction. (pg. 23)

Example:
For example, let's say that a large urban school district has observed that a small but worrisome number of new teachers resign after their first year. Records have been kept on this problem for the past 10 years, and they reveal a consistent baseline: 11, 12, 10, 10, 12, 9, 10, 11, 11, 12. Now assume that the district implements a program for all first-year teachers that offers a hotline to call with problems, pairing with an experienced teacher, and monthly therapy sessions to discuss anxieties, doubts, and other undesirable emotions.

To evaluate the effectiveness of this program after five years, the preprogram trend is compared to the following postprogram trend: 4, 5, 2, 3, 4. Further analysis reveals that such a drop in the attrition rate could hardly be explained by chance. In other words, the drop is probably real, but one must be cautious in attributing the decline to the program, for there may have been other new influences (such as a large pay increase, reduction in class size, six new schools, etc.) that could explain the resulting decline. In the absence of other explanations, the break in the trend over time corresponding to the onset of the program is fairly convincing evidence that the program was the causal mechanism.

Some of the difficulties associated with time series interpretations are revealed in Figure 10.3. The trend line in outcome A might look dramatic, but the upward slope is not likely attributable to the treatment since the change began before the treatment was implemented. Outcome B generally reveals a flat line suggesting no treatment effect. Ambiguity might arise, however, by the fact that half of the posttest measures are above all of the pretest measures (and none are below). Outcome C suggests an early treatment effect but lacks consistency in this trend over all posttest measures. Outcome D is most compelling because no apparent trend exists in the pretest measures, yet there is a clear upward trend in the posttest measures following treatment. All measures after the treatment are clearly higher than all measures before the treatment, another sign of a treatment effect.